Staff Profile Pages

Dynamics-Informed Neural Estimation of Human Joint Torques for Lower-Limb Exoskeletons

Authors: Yin, B., Zhao, T., Yu, H., Ge, J., Zhang, Y., Yuan, J.

Journal: Proceedings of Isoirs 2026 Moving Towards Embodied Intelligence in the AI Age 2026 6th International Symposium on Intelligent Robotics and Systems

Publication Date: 01/01/2026

DOI: 10.1109/ISOIRS70157.2026.11545223

Abstract:

Accurate estimation of human joint active torques (HJATs) is essential for assist-as-needed control in lower-limb rehabilitation exoskeletons. However, joint axis misalignment and soft tissue compliance introduce closed-loop rigid-flexible coupling, which degrades estimation accuracy. This paper establishes a closed-loop dynamic model of the human-exoskeleton coupled (HEC) system and derives the transmission relationship between HJATs and measurable interaction forces. Building upon this model, a dynamic-information-embedded piecewise neural network (DIE-PNN) is proposed for real-time torque estimation. Simulation results demonstrate that the proposed method improves estimation accuracy compared with a conventional RBF neural network baseline while maintaining stable prediction performance. The proposed framework provides an effective and physically interpretable solution for interaction-force-based motion intention recognition.

Source: Scopus